A Two-Stage Algorithm for Post-Nonlinear Blind Source Separation
Wai Yie Leong, John P. Homer, Zdenka Babić, Danilo P. Mandic · 2006
An approach to blind separation of post-nonlinearly mixed sources is presented. The proposed approach consists of two stages, namely the estimation of the inverse of the nonlinearity followed by standard source separation. This approach represents further proving of our previously introduced EKENS algorithm, where the critical stage of the estimation of the inverse of the nonlinearity is revised. The used of the Gram-Charlier series, makes the proposed algorithm capable of dealing with both nonlinear mappings and variations of statistical distributions of the sources. The analysis is supported by a comprehensive set of simulations which justify the proposed approach